This page covers two things: which photographs work, and what those two age fields actually control. All of it comes from testing rather than from general advice.
Choosing a photograph
- Front-facing, or close to it Profiles have a much lower success rate. The reason is on the failure cases page
- A large face A face filling a quarter of the frame or more is the safe range. When the face is small there are too few pixels for the model to rewrite
- Even light If half the face sits in shadow, the lines on that half cannot be judged
- One person in frame A second face makes it wander
- Eyes and mouth not covered Those two areas carry the most ageing information
Black and white is fine, and so are yellowed prints. The instructions state that a monochrome original stays monochrome, so nothing gets colourised behind your back.
What the two age fields control
Current age in the photo is there to work out the gap, not to check anything. Reaching 70 from 30 and reaching 70 from 60 require completely different changes. Getting this wrong throws off the size of the whole edit.
Age you want to see sets the direction and the strength. Higher than the current age means older, lower means younger. The wider the gap, the more aggressive the edit, and the easier it is to lose the sense that this is still the same person.
The lower limit is 16. That is deliberate: turning adults into children is not something this tool does.
Why the sex field changes the result
Male and female faces age along different paths, and the jaw, brow ridge and facial hair are described differently. Choosing wrong does not break anything, but the result grows in the other direction. If you are unsure, leave it unspecified and the model uses neutral wording, which is steadier but changes less.
Judging whether a photograph is sharp enough
Do not look at the file size and do not look at the pixel count. Both of those numbers lie. Look at how many pixels the face itself occupies.
The quickest test: enlarge the photograph on your phone or screen until the face fills it, then look at the corners of the eyes, the corners of the mouth and beside the nostrils. If those are already a blur, that blur is what the model sees. Ageing detail (fine lines, slackening, eye bags) lives exactly there, and with no information in those places it can only invent.
A common mistake is upscaling a small image before uploading. It does not help. Scaling spreads the same information across more pixels without creating any, and it can mislead the model about how sharp the photograph is. There are only two real fixes: find the original file, or re-shoot the print.
Three things that matter when photographing a print
If what you have is a physical photograph, the quality of your copy decides everything:
- No flash Photo paper reflects, and whatever the reflection covers is gone. Take it to a window instead and use daylight, with the light coming from the side and slightly in front
- Square on Shooting at an angle turns the photograph into a trapezoid and distorts the face with it. Hold the phone flat and shoot from directly above
- Wipe it first Dust and fingerprints become obvious once enlarged, and the model sometimes treats them as skin texture
Afterwards, your phone's photo app almost certainly has crop and auto-adjust. Cropping the background away so the face fills more of the frame is the single most useful step. Do not apply filters and do not turn on beauty mode, both of which erase the skin texture this tool needs to rewrite.
Monochrome, yellowed and creased photographs
Monochrome works and stays monochrome, because the instructions require it. Yellowing and fading do not affect the judgement either, since the model reads light and shade rather than colour.
Creases and mould marks are treated as part of the photograph and stay where they are. This tool changes age and does not restore. If a crease falls across the face, clean that area up in your phone's editor before uploading.
What a given gap actually looks like
These come from the 1863 Lincoln portrait: one source photograph, one set of settings, only the target age changed. The images are on the examples page.
- 10 to 15 years The beard and temples turn grey and fine lines appear around the eyes. Features and bone structure do not move at all, so it reads as the same person a few years on
- 20 years Deeper lines, whiter hair, eyelids starting to drop. Clothing and background are still as they were
- 30 years Deep lines across the face, slackening at the jaw and neck, hair almost fully white. Still recognisable at this size of change
- 40 years and more Lines, slackening and white hair all at once. Eye shape and face shape usually hold, but this is where surprises happen. Confirm the photograph works at a smaller gap before jumping up here
Going younger follows the same pattern in reverse. Only past a 20 year gap does the model genuinely repaint the skin; below that it just makes the nasolabial folds fainter.
Common kinds of photograph
- Passport and ID photographs The best case. Front-facing, evenly lit, large face, all four conditions met at once. The background is usually flat colour and the model leaves it alone, which is what you want
- Family groups Crop to one person first. With two faces in frame the model does not know which to change, and it usually changes both or invents a third
- Wedding and studio portraits Usually retouched already. Skin smoothed until there is no texture leaves nothing to rewrite, and the aged version looks like wrinkles stuck on top. These work better going younger
- Photographs of a screen These carry moiré patterns, and the model sometimes treats the banding as skin texture. Use the original file if you can find it
- Family members who have died Nothing stops you, and this is one of the most common uses. The one thing worth saying is that the result is the model's guess, not what that person would actually have looked like
How to try it the first time
Start with a small gap (say 40 now, 55 wanted) to confirm the model has a grip on this particular photograph, then widen it. Jumping straight to a 40 year gap leaves you unable to tell whether an odd result came from the photograph or from the size of the change.
The same photograph with the same settings also varies between runs, because the random seed differs. If the expression or some detail is off, running it again usually fixes it, and you do not need to touch the photograph first.